2004
tem
use
The
its
etic
ates
our
it is
to
are
TE
one
lon
the
rth
jor
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
3.2 Data Resources DEM.
Many kinds of data were used in this experiment. They were
remote sensing images. topographie maps. land use maps. and
32.] Remote Sensing Images:
Type Number Phase Resolution Bands
TM 1Scene 2001/09/01 30m TM1,2,3,4,5,7
as 1Scene 2002/10, 2003/10 ;
SPOT4 10m panchromatic
1Scene 2002/10, 2003/10
multi-spectral,
SPOTS 1Scene 2002/10. 2003/11] 108-2. 507].— "tts epectra
panchromatic
Arial photos 16sheets 2003/1 1m RGB color
Table 3. RS Images Used in Jinnan Experiment
3.2.2 Maps:
1:10000 topographic maps: 192 sheets used to rectify SPOTS
(2.5m) images
1:50000 topographic maps: 32 sheets used to rectify SPOTS
(10m) images
1:10000 land use maps: 28 sheets used to change detection and
map updating
3.2.3 Dem:
1:50000: 32 sheets used to otho-rectify SPOTS (2.5m) images
1:230000: 32 sheets used to otho-rectify SPOT4 (10m) images
3.3 Experiment Content
(1) Using middle resolution images (30m landsat7 and 10m
SPOT4) to update land use maps made in several years ago. (2)
Using high resolution images (SPOTS 2.5m and Im aerial
photos) to update old land use map.
34 Result
[Lis no meaning to statistic interpretation precision. because the
precision of change spots area and their border points!
coordination is not satisfied with the demands of 1:10000 map
scale accuracy extracted by middle resolution images (30m
landsat7 and 10m SPOT4). and other assistant materials to help
us judging more detailed change information could not be
gathered. many change spots during past years can not be
detected and extracted. So. the test of land use investigation
with two phases of middle resolution images is failing.
Investigation with one phase of high resolution images and
other phase of middle resolution images was this experiments
emphases. From this experiment. some conclusion was
obtained.
341 Extraction accuracy of Land use Category: Eleven
categories of land use were extracted by interpreting from
SPOTS (2.5m) and checked by navigation GPS out of the door.
We found that orchard. water. river and beach can be extracted
correctly. But forest and: waster land were identified not so
good. The reason is that waster land is very easy to be confused
311
with breed-pond. and forest is casy to be confused with weed
by not only computer but also human interpretation since they
have the similar spectral characteristic in images.
We compared the spots category identified by RS images in
doors with its real category investigated at the real place, and
got the extraction accuracy of several typical land use category
which is shown at the following table.
Land Use Identified in Doors Right Percent
Category Wrong Right (10094)
F number number
Infield 5 34 89.74
Orchard 0 62 100
Forest 10 15 60
Water 0 88 100
Town 2 2 100
Village 5 83 ; 92
Factory 1 20 91
Road l 4 100
Waster Land 1 3 75
River 0 2 100
Beach 0 8 100
lable 4 Land use Category Extraction Accuracy
3.4.2 Border Points’ Coordinate Precision: Change spots’
border was surveyed with GPS. Compared to GPS data. the
least RMS of spots border of images is 7.44m. and the biggest
RMS of spots border is 9.86m. That is according with the need
of scale 1:10000 maps.
